Anthropic has rolled out a dedicated mobile application, Claude Duet. This personal AI agent is designed to proactively distill key information from users' linked apps and manage tasks like payment reminders, contract drafting, and order placements on behalf of users. Operating on a cloud-based virtual machine, it facilitates multi-device access, although its availability is currently restricted to a limited number of paid subscribers in the United States. Lately, a number of firms, such as Meta and OpenAI, have also unveiled comparable offerings. These products typically hinge on cloud-based functionality, gaining control over users' email, calendars, and even payment authorizations. Nevertheless, the widespread adoption of personal AI agents encounters several hurdles, notably users' apprehensions regarding security concerns like permission allocation, data breaches, and prompt injection attacks. Numerous online platforms, guided by their own policies and interests, outright prohibit AI agent access. Some users have even faced account suspensions for policy breaches, merely for utilizing AI agents to peruse content. At present, the industry is pursuing two strategies to overcome these challenges: one involves having AI agents mimic human actions to circumvent platform limitations; the other entails employing specialized protocols, such as MCP, to enable platforms to recognize AI agents and set up transparent permission guidelines for seamless interaction between the two entities.